ODE | Neural Ordinary Differential Equations - Best Paper Awards NeurIPS

[@nathancooper1001]

Exciting work! I look forward to to being able to eventually understand it haha

[@TheSam1902]

This video is amazing. Researching this very paper since last week and seeing the actual presentation is lifesaving

[@declan6052]

Wow, at 9:59 we see the very beginning of diffusion models! Very cool to see something so impactful in its infancy

[@rstar0000]

Amazing work!

[@DrOsbert]

Please do support our channel by subscribing and sharing the video to others. Love & Peace!

[@dustinkendall3365]

So do these odes model a function as simply a function of time, because I want to make it as a function of time with position and velocity as initial condition parameters. How do I do this?

[@rickymort135]

well it's ODEs of the form x' = f(x, t) so f is modelling the velocity - the only inputs you can put in are positions and time. To input velocity you'd need f to give you: f(initial_position) = initial_velocity. This is a little tricky. You could include include training examples [initial_position, initial_velocity] and train f on this with a big penalty for deviation

[@bananarobotoverlord]

So 1/2 the parameters, but 2-4x the FLOPs?

[@DrOsbert]

You can say so but It depends on the actual use case

[@ranam]

2:26

[@OneShot_cest_mieux]

What is theta please ? I don't understood.

[@youcancode6988]

Theta is the parameters

[@OneShot_cest_mieux]

@@youcancode6988 What parameter for example ? And theta seems to be a function of time

[@OneShot_cest_mieux]

@@youcancode6988 I hope you don't comment your code by writing "var1 is the first parameter, var2 is the second parameter, [...]"

[@youcancode6988]

@@OneShot_cest_mieux Theta is the weighted values given to each input on each node in each layer of the neural network

[@OneShot_cest_mieux]

@@youcancode6988 ok thank you

[@iworeushankaonce]

Amazing talk, and a huge dislike for the amount of commercials you put in this video. shame